porter/docs/developing/test-autoscaling.md

Test Cluster and HPA Autoscaling

Prerequisites:

• metrics-server must be installed (installed by default on all Porter clusters).
• cluster-autoscaler must be enabled (enabled by default on all Porter clusters).
• Have kubectl access to the cluster, as it is easiest to port-forward to the exposed service.
• Download hey or an equivalent.

Steps

1. Launch a Docker template with image URL k8s.gcr.io/hpa-example. Enable autoscaling under the Resources tab -- suggested params are 100 Mi RAM, 100m CPU, 1-100 replicas, 50% CPU & RAM util. Do not Expose to external traffic to run load tests.

2. Confirm that the horizontal pod autoscaler was created via kubectl (kubectl get hpa) and that the metrics server is reporting metrics via kubectl top pods. Current utilization may initially show up as <undefined>, but this information should load after a minute or so.

3. Port-forward to the service via kubectl port-forward svc/<svc-name> 10000:80.

4. In a different terminal window, run hey http://localhost:10000. Vary the hey parameters to test autoscaling under different loads (hey -h for options).

Check that replicas scale when passing the threshold from the metrics tab. Use kubectl to probe the nodes and confirm cluster-level upscaling has also occurred. Delete the chart and confirm node downscaling when done.